154 / 168 · 01 Agent Skills for Browser Automation · Decision Framework: When to Use What (July 2026)← prev⊞ allnext →☰ Read as one page
21.5Risk Assessment
| Risk | Playwright CLI/Agents | Playwright MCP | Stagehand | browser-use | Vibium | Selenium |
|---|---|---|---|---|---|---|
| Vendor lock-in | Medium (Microsoft) | Medium (Microsoft) | Medium (Browserbase) | Low (OSS) | Low (Apache 2.0) | Low (W3C) |
| Tool discontinuation | Very low | Low | Medium (startup-backed) | Medium (community) | Medium-high (no adoption yet) | Very low (20+ years) |
| Token cost scaling | Low | High | Medium | High (vision) | Low | N/A (not AI) |
| Maturity gaps | Few (new CLI on mature engine) | Few | Some | Some | Many (V1) | Few |
A January 2026 lesson worth internalizing: Octomind — a credible, recommended tool six months ago — is winding down. In this space, tool-discontinuation risk is not hypothetical; architect so that your test plans (Markdown, portable) outlive any particular execution tool.